Jerusalem's Lightricks releases its 4K video-and-sound AI model with open weights

On 6 January 2026, at the CES technology show in Las Vegas, the Israeli company Lightricks made the full LTX-2 model public, including its weights, inference pipelines and training code. LTX-2 generates video and matching sound together, with synchronised speech and lip movement, producing clips of up to about 20 seconds at native 4K resolution and up to 50 frames per second. The release was announced alongside Nvidia, whose role was to optimise the model for its graphics cards: a compressed version cuts its size by roughly 30 percent and runs up to twice as fast on consumer RTX GPUs, while the same model scales up to data-centre systems.

The move set Lightricks apart from the large American labs, whose leading video generators, such as OpenAI's Sora and Google's Veo, are available only as closed cloud services. Because the weights are public, users can run LTX-2 on their own computers, keeping private material off outside servers, and can retrain it on their own footage to create custom visual and sound styles. The Israeli tech site Techtime described the decision as an attempt to make the model a standard that others build on, with the company earning money from tools and services rather than from the model itself. The license is free for academic research and for companies with less than $10 million in annual recurring revenue; larger businesses need a commercial license. Lightricks followed up in March with LTX-2.3 and a desktop editor that runs the model locally.

Lightricks was founded in Jerusalem in 2013 by a group that included researchers from the Hebrew University of Jerusalem, and it became known for the selfie-editing app Facetune before turning to generative AI with its LTX Studio platform. Chief executive and co-founder Zeev Farbman called LTX-2 the first truly open audio-video model.